a) Global effects
We first analyzed global connectivity measures using group t
test and Pearson correlation with clinical measures for the patient group. For mean fractional anisotropy (FA) values, the between-group difference was significant at p<0.0001, the number of fibers detected by fiber tracking algorithm differed at p<0.03, but the total length of detected fibers was not different between groups. The mean value of DTI connectivity (averaged over whole 6000×6000 matrix) was lower in patients than controls (p<0.005). For patients this measure also significantly correlated with the positive symptom score of the Positive and Negative Symptom Scale (PANSS) (p<0.002), and with Thought Disorder Index (TDI) (36
) scores (p<0.02), but not with general or negative PANSS scores. More specific analysis of positive PANSS subscores showed that global DTI measures correlated negatively with indices of social avoidance (r= −0.57, p<0.001) and hallucinations (r
= −0.45, p<0.01).
The global average of resting connectivity showed no significant difference between groups, but was positively correlated with positive PANSS (p<0.01) and TDI (p<0.005) scores and negatively correlated with the negative PANSS score (p<0.01).
In agreement with prior results (5
), the spatial correlation between global arrays of resting and DTI connection was significant in the control group, mean r
= 0.16±0.02 (p<0.0001), and weaker but still significant in patients r
= 0.13±0.04 (p<0.002).
In a between-group comparison, the correlation between anatomical and functional connectivity matrices was significantly lower in patients than controls (p<0.02). The correlation of this measure with symptom scores in patients revealed a negative correlation with positive (p<0.05) and general (p<0.02) PANSS scores, but a positive correlation (in contrast to trend suggested by the group difference) with positive PANSS (p<0.05) and TDI scores at (p<0.05).
b) Regional Correlation maps
To further analyze and localize the spatial coherence between resting and DTI connectivity we calculated, for each of 28 predefined ROIs, resting and DTI connectivity map showing for every voxel the strength of its connectivity to this given ROI. Thus for each ROI we were able to quantify the spatial coherence between spatial and functional connectivity of network originating in it. All regions showed significant coherence between both connectivity measures in control subjects, and albeit less significantly for patients. These regional values were then compared between groups using t tests. For one region, posterior cingulate, the between-group difference in this between-modality coherence value was significant (p<0.02), with patients showing less coherence. Thus, in the patient group the functional connectivity pattern originating from posterior cingulate was less similar to this region’s pattern of anatomical connections in controls, suggesting that this network may be the most affected by functional reorganization related to schizophrenia.
c) Interregional correlations
For every subject and each of 28×27/2 region pairs, we analyzed the mean resting and DTI connectivity averaged for all voxel pairs connecting chosen regions and compared values between groups. For each connection between to ROIs and for each connectivity measure (anatomical and functional) we calculated two statistical maps, one of the significance of the between group difference , and the other of the correlation between connectivity measure and positive PANSS score in the patient group.
To account for multiple comparisons we used logical combination of those maps. The connection was considered to be affected when it showed a significant group difference and a correlation between connectivity measure and clinical symptoms. The differences in anatomical connectivity were more robust and we used threshold of p<0.01 for such comparisons, leading to p<0.04 after correction for multiple comparisons. No changes in functional connectivity survived at this significance level; thus we defined as affected all connection pairs that were significant at p<0.02 for the group difference and correlation comparison, and were also characterized by significantly lower anatomical connectivity at p<0.01.
Anatomical connectivity strength among 12 region pairs differed significantly between groups. For all such pairs the anatomical (DTI) connectivity was lower for schizophrenia patients and correlated negatively with intensity of positive PANSS measures. These connections ( and ) originated in lingual and cingulate gyri and included connections with left inferior parietal lobule, middle and superior temporal gyri and inferior frontal gyrus.
Figure 1 Connections between brain regions that differ between schizophrenia patients and controls in measure of anatomical (measured with DTI) connectivity (). Only connections for which anatomical connectivity was lower in patients than control were significant (more ...)
Table 1 Region pairs for which there is significant difference in anatomical connectivity. Anatomical connectivity is lower in patients and negatively correlated with positive PANSS subscores for all region pairs. The anatomical position of those connections (more ...)
Four of the region pairs showed a between-group difference in strength of functional connectivity ( and ). In schizophrenia, all such connections showed significantly lower anatomical connectivity but three showed higher functional connectivity. These latter linked ventral anterior cingulate gyrus and ventral medial frontal gyrus with left inferior frontal gyrus, as well as thalamus with left postcentral cortex. Only one connection, linking left middle temporal gyrus with postcentral gyrus, showed lesser functional connectivity in schizophrenia.
Figure 2 Connections between brain regions that differ between schizophrenia patients and controls in measure of functional connectivity (). Red lines represent connections for which the functional connectivity was higher in patients, while for green line (more ...)
Table 2 Region pairs showing significant difference in functional connectivity (measured by resting correlations). Anatomical connectivity is lower in patients and negatively correlated with positive PANSS subscores for all region pairs. The anatomical position (more ...)
d) Network analysis
K-means clustering algorithm led to delineation of the Task Positive Network (TPN) and default Mode Network (DMN) with DMN later subdivided into two components that exhibited different group differences. The between-group analysis mean connectivity averaged over all connections within those components showed both significantly less anatomical connectivity within each component separately and also between component connections in schizophrenia. Mean functional connectivity differed significantly only in TPN and was lower in patients.
An additional K-means analysis was performed on each component defined above using the anatomical connectivity data averaged separately for control and patient groups. This led to a subdivision of DMN two subcomponents as shown in .
Figure 3 Components detected by k-means cluster analysis of connectivity. The functional resting connectivity led to identification of the default mode network (DMN), depicted in red and orange and its counter part – the task-positive network shown in (more ...)
shows the group comparison and correlation to clinical scores in schizophrenia patients for the internal connectivity within each component and between DMN and task positive network as well as between positive and negative subcomponent of DMN.
Table 3 This shows differences between control and patients in the internal connectivity for each of K-means defined components. Correlations with clinical measures of positive PANSS subscores and Thought Disorder Index also presented. Same data presented for (more ...)
Two DMN subcomponents showed striking differences in between-group comparisons. DMN-1, that included anterior cingulate and portions of posterior cingulate cortex (PCC), showed higher functional connectivity in patients and commensurate positive correlation of functional connectivity with clinical symptoms, although no between-group differences in strength of anatomical connectivity. DMN-2 included parts of bilateral parietal cortices and bilateral dorsolateral prefrontal cortex (DLPFC) and was characterized by lower anatomical connectivity in schizophrenia. The values for subcomponents of DMN in patients and controls are presented in .
Figure 4 Mean functional (left column) and anatomical (right column) connectivity averaged over all voxel pairs within the DMN-1 (top row) and DMN-2 (bottom row) subcomponent of Default Mode Network. Green crosses represent values for healthy controls. Only voxel (more ...)
The analysis of spatial coherence between functional and structural connectivity presented earlier for global connectivity matrices was extended to subnetworks DMN-1, DMN-2 and TPN. Only TPN (p<0.001) and DMN-1 (p<0.01) showed significant between-group differences, with lower coherence in schizophrenia patients. For those two subnetworks, the correlation with positive and general PANSS measures were significant at p<0.05, consistent with between-group differences (i.e. coherence decreased with increasing symptom severity). The correlation with TDI was significant for both TPN (p<0.02) and DMN-1 (p<0.01), but surprisingly for TPN the direction of difference in TDI was opposite to that of the PANSS, with the between group comparison for coherence increasing with increased TDI.
e) Results summary
The global tests of mean connectivity showed that schizophrenia patients had decreased anatomical connectivity and caused decoupling between anatomical and functional connectivity. Those findings were confirmed by inter-regional connectivity analysis and regional maps, localizing the decoupling to networks originating in PCC. Further network analysis showed that schizophrenia-related alterations affected portions of the DMN.